What Can I Undo If I Build an AI Agent and Change My Mind?

You're partway through a build and a doubt shows up. Did you give it too much? Can you take it back? On October 7, 2026 I showed or described each of the controls below, and the documentation adds a few details.

Can I change what I told the AI agent?

Yes. After the AI agent wrote its own instructions, I said all of it is editable and named the easiest place to do it:

From the recording, 42:55. So all of this is editable. The easiest way to edit would be inside of the chat bot.

At 43:03 I added that the chat feature is the easiest way to edit both skills and instructions.

An edit you haven't saved yet has its own undo. Gumloop's documentation says nothing saves until you press Save, and that the arrow next to Save undoes unsaved changes.

What if the AI agent edited its own instructions?

That's the one place I'd slow down. I pointed out the switch, a blue toggle button that was on:

From the recording, 24:18. You'll notice there's something that says edit own instructions with a blue toggle button that is on.

I keep a 15 minute review on my calendar every three weeks to read the instructions (24:51), because the AI agent updates them automatically. The documentation says the AI agent changes its instructions once you correct it, that there's no version history, and that you revert by editing the prompt manually. So the review is how you catch a change you didn't want. Checked on 2026-11-05 against Gumloop's Agents documentation. How Often Should You Review an AI Agent's Instructions? covers the cadence.

Can I take back a connector's access?

Yes, for the connector I showed. Under connectors, on the far right of the Gmail row, there was a circle with a check mark and three dots. The three dots open a menu that switches the account or removes access (45:04). I removed my Gmail access on the spot:

From the recording, 45:18. So just like that, my Gmail access has been removed. I'm going to re-add it for the purpose of demo, but I wanted to demonstrate how you remove access.

I re-added it for the demo. How Do You Control What an AI Agent Can Do in Your Gmail? has the full screen.

Can I stop a single tool without removing the connector?

Two ways. In the first, for a tool you haven't connected, you say so in the chat. I had planned to use Slack, then changed my mind:

From the recording, 38:16. I can now tell the agent I'll pause on that so it won't continue to prompt me to ask for that.

In the second, you use the tool settings under the connector. At 44:19 I showed that I can give permission or restrict access to sending emails, updating, creating drafts and forwarding emails. Gumloop's documentation describes four presets for a connector's tools: always allow, ask each time, ask for writes and deletes, and custom, where each tool can be allowed, asked about or denied.

Can I delete the AI agent?

Yes, and I said it's as simple as creating one:

From the recording, 48:45. So just as simple as it was to create, they make it equally simple to delete.

In the session I described where the delete sits: three dots next to the box with an up arrow and a half open box (47:56). If you see an error, I said to refresh first, because sometimes the delete already went through (48:16). The documentation lists two places to delete: the Destructive section on the Settings tab, and the three-dot menu on its Agents page. It says the menu in the chat header doesn't have Delete, so use the two places it lists, and it says a confirmation dialog lists what the delete affects. If the dialog shows something, you type Delete agent to confirm. It also says the AI agent can't delete itself or turn off its own tool approvals.

Neither my session nor the page I read says a deleted AI agent can be restored, or what happens to stored data after a delete. I can't tell you either one.

Checked on 2026-11-05 against Gumloop's Agents documentation.

On my relationship to Gumloop, I said this:

From the recording, 33:22. I don't work for Gumloop. I found this tool over the summer through Lenny's podcast, and I started using it and have since made myself an ambassador for them. I've created a relationship with Gumloop. I am not an employee.

More on that answer: Do you work for Gumloop? (the answer from the session).

What does the file keep from happening in the first place?

The file on the session page has a never list for the Chief of Staff AI agent. Deleting is in it:

From the file on the session page. You never delete anything, in my inbox or my calendar.

The file also says every reply stays a draft until the words "send it" come in that exact turn. That's a written rule in a file, and a file isn't a lock, so the tool settings above still matter. How Do You Delete an AI Agent? shows the delete screen as I recorded it.

What goes wrong

An edit nobody read. I left the self-edit toggle on, and at 43:10 I told the group to put a reminder on their calendar to go in and read the instructions.

An error when you delete. I said to refresh first, because sometimes the delete already went through (48:16).

What the recording doesn't show: the contents of the delete confirmation, or what the AI agent had written in its instructions before I edited them.

Where this comes from

This article grows from the session I taught live with Women Applying AI on October 7, 2026. The quoted passages from the recording are my own words, taken verbatim. The passage marked as from the file on the session page comes from the files published with that session. The tables, steps and checklists around them were drafted by my team's AI agents from those words and my published pages, and checked against the recording before publishing. The file is on the session page. Platform facts that have a public source carry the dated line above, and screens change.

Build Your First AI Agent in One Hour: the full session with the clips, the prompts and the transcript.

Related reading: How Do You Delete an AI Agent?, How Do You Control What an AI Agent Can Do in Your Gmail?, What Are AI Agent Guardrails?.